---
name: undominated-safety-refusal-calibrator
description: Calibrates safety classifier thresholds to minimize benign query refusals while maintaining strict compliance boundaries.
license: MIT
metadata:
  author: Undominated.ai
  version: "1.0.0"
  tags: ["safety", "refusal", "calibration", "compliance", "developer-experience"]
---

# Safety Refusal Calibrator

## Mission & Boundary
Calibrates safety classifier thresholds to minimize benign query refusals while maintaining strict compliance boundaries.

## When to Deploy
Deploy when developer teams experience over-refusal friction on technical or domain-specific workflows.

## Operational Boundaries & Limitations
Does not loosen boundaries on core safety violations (CSAM, CBRN, self-harm). Calibrates benign edge cases.

## Execution Protocol
Review false-refusal logs, evaluate classifier confidence margins, and tune system prompt disambiguation instructions.
